Spectral measuring, or analytical instruments, like Near Infrared Reflection (NIR) and Near Infrared Transmission (NIT) instruments, are widely used in e.g. the food industry or in agriculture to analyse the composition and nutritive value of foods and of crops such as forages. For example, the amount of constituents such as crude protein, fat, and carbohydrates may be determined by spectral analysis.
Direct analyses of e.g. a sample of rice on two spectral analysis instruments of the same design will probably produce somewhat different results. The differences may be due to manufacturing variability and instability of the instruments.
In order to obtain comparable results when analysing the same sample on several spectral analyzers of the same design, the instruments are standardized. Thus, after standardization, the analytical results from each of several “satellite” instruments, i.e. instruments located at different measurement sites, are corrected so that they correspond to the analytical results from a “master” instrument.
In this standardization procedure, a so called “standardization cup” containing a reference material is usually used. This standardization cup is hermetically sealed so that the quality of the reference material is maintained over a long period of time. Firstly, the reference material is analysed using the master instrument. The result from this analysis, in the form of a “master” spectrum of the reference material, is stored on a disc together with information identifying the reference material and information identifying the master instrument.
Thereafter, this disc and the standardization cup are sent to a user of a satellite instrument. The standardization cup is then put into the satellite instrument and the disc is inserted into a computer connected to the satellite instrument. The reference material is then analysed by the satellite instrument whereby a “satellite” spectrum for the sample is produced. This satellite spectrum is then compared to the master spectrum on the disc by means of a software program on the computer. If the satellite spectrum differs from the master spectrum, which is usually the case, the program produces a standardization model which mathematically transforms the satellite spectrum to correspond to the master spectrum. The standardization model is then stored on the computer and may be used to transform the results of future routine analyses on the satellite instrument for samples of the same type of material as the reference material.
A problem with this known standardization procedure is that there is always a risk that a disc with “standardization” information relating to a specific reference material is lost, or worse, is confused with some other standardization disc for another reference material. The consequences of lost standardization discs are delayed or cancelled standardizations. The consequence of confusion as to the identity of standardization discs is error in predicted values for all routine samples measured on that satellite instrument, at least until the next time the satellite instrument is standardized.
For a standard routine analysis, where the composition of an unknown sample (e.g. grain from a process line) is to be analysed, an operator puts the sample into a sample cup and the sample cup into the (satellite) instrument. Then the operator enters information about the sample type (grain etc.) on the computer, chooses from a data base on the computer an appropriate standardization model and a prediction model to be applied to the sample and starts the analysis of the sample by the instrument. During the analysis, a spectrum of the sample material is collected, standardized and has a prediction model applied to it in order to translate this standardized spectrum to predictions of chemical composition. It is desirable that routine analysis of samples is more automated, so that mistakes due to “the human factor” may be eliminated and the procedure may be quicker. With the existing system there is always a risk of having erroneous prediction values if the operator chooses the wrong standardization model or prediction model.